AGP 8X bandwidth VS. Graphic card memory bandwidth

[Grishnakh]

Hello, I'm new to Anandtech.
I have a question about the title mensioned:

I know the AGP bandwidth count small on performance, but I don't know why.
Why a AGP 8X (2.133GB/s) bandwidth can fulfill the need of Graphic card,
even in monsterous Geforce 6800Ultra with memory bandwidth of 35.2GB/s

I guess that these 2 bandwidth have little relationship, but how should we evaluate
the exact AGP bandwidth that can feed a certain Graphic card.
(That is, for example, how much AGP bandwidth do a Geforce 6800Ultra need?)

 

[Cat]

AGP bandwidth is used to send and receive data to and from the CPU. This is pretty slow, as you've already noted.

The quoted memory bandwidth you see refers to the amount of data the card can push through itself, once that data is onboard, and is mostly taken up by fill-rate (pixels being drawn.)

 

[imported_Onizuka]

Slow....it's not even being used to full capacity, not even at 4x. Why? Because I change my AGP to 4x from 8x on a FX5900 and observe no difference in speed whatsoever. I'm running an Athlon64 3400+ speed too.

 

[Cat]

There are numerous possible bottlenecks in the applications you run. Just because you doubled your AGP bus speed does NOT mean that it's not being used to full capacity. It merely means that it's not the bottleneck. If you're playing games, you are most likely NOT transform limited, but CPU or fill-rate bound.

 

[Lonyo]

The graphics card needs data, this can be supplied easily by the AGP slot, even at 4x.
The graphics card manipulates that data over and over again, and does lots of operations on it.
Since the data is stored in the RAM on the graphics card, it needs to be supplied really quickly, this is why you need lots of memory bandwidth.
The data sent to the card isn't as much as what is used by the card (it's the same data, just used over and over on the card - so there's more if you see what I mean?)